I'm writing a Bachelor Thesis in engineering an a small part of it has two deal with assigning weights to independent measurements. This is from the thesis:

I am pretty sure that what equations (3.23) and (3.24) are correct, the fit my data perfectly, but I can't for the life of me find a source to confirm them. I've found two Wikipedia articles relating to them, but they don't really give any sources. Does anyone know whether what I wrote is correct, and if so, where I could find a source to cite?

http://en.wikipedia.org/wiki/Inverse-variance_weighting http://en.wikipedia.org/wiki/Weighted_arithmetic_mean#Dealing_with_variance

  • $\begingroup$ This is often (incorrectly, in my opinion) referred to simply as Weighted Least Squares, and a search for that term will find you plenty of sources. $\endgroup$ – shadowtalker Sep 17 '14 at 14:39
  • $\begingroup$ This result is demonstrated in many threads on this site. To expedite my search I limited it to threads where I posted answers and found stats.stackexchange.com/questions/5392, stats.stackexchange.com/questions/43159, and stats.stackexchange.com/questions/12251. The latter is the most directly relevant one. It corrects two misconceptions apparent in the thesis: (1) it explains how this is the minimum variance unbiased estimator and (2) no assumption of Normality is needed. $\endgroup$ – whuber Sep 17 '14 at 16:03
  • $\begingroup$ Thanks for the answers. They did make this quite clear. My main problem is still that I can't find a proper source to cite in my thesis. Does anyone know of a statistics book or an article which clearly states that (3.24) is the minimum variance unbiased estimator? $\endgroup$ – Gunnar Sep 20 '14 at 14:43
  • $\begingroup$ No source needed, I'll just derive 3.24 from 3.23 with the method of Lagrange multipliers $\endgroup$ – Gunnar Sep 20 '14 at 16:13

For the case of two weights, you can simply set $w_2 = 1 - w_1$, plug it into (3.23) and then find the minimum of the quadratic function of $w_1$.

In general, as explained here on p 5, you have to apply the Lagrange multiplier method that will give you a closed-form solution unless you restrict the weights to be non-negative.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.